|Molecular Vision 2004;
Received 8 March 2004 | Accepted 10 June 2004 | Published 15 June 2004
The surface of visual arrestin that binds to rhodopsin
W. Clay Smith,1,2 Astra
Dinculescu,1 James J. Peterson,3 J. Hugh McDowell1
Departments of 1Ophthalmology and 2Neuroscience, University of Florida, Gainesville, FL; 3Department of Biology, Kansas State University, Manhattan, KS
Correspondence to: Dr. W. Clay Smith, Department of Ophthalmology, Box 100284, JHMHC, Gainesville, FL, 32610-0284; Phone: (352) 392-0476; FAX: (352) 392-0573; email: firstname.lastname@example.org
Purpose: The binding of visual arrestin to phosphorylated, activated rhodopsin serves as a model for studying the inactivation process of a large class of G-protein coupled receptor systems. In this study, we combine the use of insertional mutagenesis, fluorescence labeling, and scanning alanine mutagenesis to identify the surface of interaction between arrestin and rhodopsin.
Methods: The ten amino acid myc tag (EQKLISEEDL) was inserted in eleven loop structures that connect βstrands and the tagged arrestins were heterologously expressed in yeast. Binding competition assays were performed with these proteins, using an anti-myc monoclonal antibody. Site specific cysteines were also substituted in selected loop structures in arrestin. These cysteines were labeled with a fluorescent reporter to assess the proximity of the introduced cysteine with rhodopsin in the bound complex.
Results: Competitive inhibition of arrestin binding to light activated, phosphorylated rhodopsin with an anti-myc antibody showed that all competitive sites lay along a single surface encompassing the N- and C-terminal domains. Fluorescence labeling of these loop structures and subsequent interaction with rhodopsin indicates close apposition of loops 68-78 and 248-253 to rhodopsin in the receptor bound state. Scanning mutagenesis of loop 248-253 implicates Ser-251 and/or Ser-252 as a potential interaction point with rhodopsin.
Conclusions: Our results clearly suggest a surface of arrestin to which rhodopsin binds upon light activation and phosphorylation. This surface encompasses elements from both the N- and C-terminal domains of arrestin.
The interaction of proteins forms the basis for most biological signaling pathways. These interactions have evolved to meet the requirements of low background noise, appropriate amplification, and duration of action for signal detection. The mechanisms regulating these protein-protein interactions are diverse, and include small ligand or regulatory subunit binding/release, protein conformational changes, and post-translational modifications such as phosphorylation.
Phototransduction has long served as a model system for G-protein coupled receptor signaling cascades, and utilizes many of the above processes to regulate the protein-protein interactions of the cascade [1,2]. For example, the interaction between rhodopsin (a G-protein coupled receptor) and transducin (a heterotrimeric guanine nucleotide binding protein) is regulated by a conformational change in rhodopsin induced by the absorption of a photon. Interaction of transducin with cGMP phosphodiesterase is regulated by a two step process involving the exchange of a nucleotide cofactor on Gα (GTP is exchanged for GDP) and release of an inhibitory or blocking interaction with Gβγ. Quenching of rhodopsin's ability to activate transducin also involves multiple steps. First, photoactivated rhodopsin (metarhodopsin) is phosphorylated on its C-terminus at multiple serine and threonine residues by rhodopsin kinase. The affinity of phosphorylated metarhodopsin for transducin is weakened but quenching is incomplete until final binding of a regulatory protein, arrestin. This last step is thought to sterically block G-protein from interacting with rhodopsin though this has not been directly shown. The speed of arrestin binding to rhodopsin is several hundred times slower than that of G protein, permitting a measured duration of signal amplification before signal "arrest" .
Though it has long been known that binding of arrestin requires both metarhodopsin II formation and phosphorylation, the detailed mechanism by which arrestin attains its specificity for phospho-metarhodopsin (R*P) is not fully delineated. A model has been proposed whereby arrestin is maintained in an inactive conformation through an intramolecular salt bridge network . These interactions assure quiescence that is disrupted by the phosphates on rhodopsin's C-terminus. This then allows arrestin to adopt the conformation that can bind metarhodopsin II. There are several lines of evidence that support this activation model. First, Schleicher et al.  showed that there was a high Arrhenius activation energy (165 kJ/mol) associated with arrestin binding to phosphorylated metarhodopsin II. This activation energy may be caused by conformational changes in arrestin during binding. Second, binding to phosphorylated rhodopsin is accompanied by regional changes in arrestin's susceptibility to limited trypsinolysis . Similar changes in trypsin susceptibility also occur in the presence of the polyanion heparin  or a phosphorylated peptide that mimics the fully phosphorylated rhodopsin 19 amino acid C-terminus (7-PP) .
The sites of interaction between arrestin and R*P in the bound state have not been fully identified. Using scanning mutagenesis, key residues in the cytoplasmic loops of rhodopsin have been identified as binding to arrestin [8,9]. Similarly, the 68-78 loop of bovine visual arrestin was shown to be involved in binding to R*P . Other studies have suggested that the binding matrix on arrestin also includes residues in the pockets of the N-domain and C-domain of arrestin [4,11-14]. Although there is evidence that arrestin undergoes large conformational changes during activation which might bring these residues together, small angle x-ray diffraction studies suggest that arrestin binding may involve smaller changes, primarily changes in the more flexible loop structures [15,16]. In this study, we use these flexible loops to empirically identify the surface of arrestin that binds to R*P in the receptor bound complex. This analysis is important not only for the study of arrestin/rhodopsin interaction, but also for studying the interaction of the nonvisual arrestins with other G-protein coupled receptors since the crystal structures show visual arrestin and β-arrestin to be highly homologous .
The cDNA coding for the ten amino acid myc-tag (EQKLISEEDL) was inserted at positions Arg-37/Val-38, Ile-72/Asp-73, Ala-99/Thr-100, Asp-162/Lys-163, Pro-186/Gln-187, Ser-199/Asp-200, Tyr-250/Ser-251, Glu-314/Gly-315, Leu-338/Leu339, Thr-343/Ser-344, and after Glu-399 using overlapping PCR products in the bovine visual arrestin cDNA. For example, for insertion of the myc tag between Tyr-250 and Ser-251, the 5' portion of the arrestin cDNA was amplified with primers that matched the 5' end of the arrestin cDNA paired with a primer that matched the codons for amino acids 246-250 and included the cDNA for the myc tag. Similarly, the 3' portion of the arrestin cDNA was amplified with a primer that had the myc codons added to the 5' end of the codons for amino acids 251-255 and paired with a primer matching the 3' end of the arrestin cDNA. The resultant two PCR products were then mixed, denatured, and allowed to anneal at the myc-tag overlap. These overlapping products were amplified using the 5' and 3' primers for arrestin containing EcoRI sites, and cloned into the Pichia shuttle vector for heterologous expression in Pichia pastoris as previously described [10,17]. All insertions were confirmed by DNA sequencing.
Alanine and cysteine substitutions were made in the arrestin cDNA by mega-primer PCR , as previously described . In all cases, the arrestin cDNA included an N-terminal (His)6 tag between the initiating methionine and the second amino acid (i.e., MHH HHH HKA NK...EAA MDE). Addition of this tag has no detectable effect on the selectivity of arrestin for R*P in our binding assay (results not shown). The following substitutions were made: I72C, S156C/S199C, V177C, S251C, V248A, L249A, Y250A, S251A, S252A, D253A, and Y254A. The mutated cDNAs were prepared with flanking EcoRI sites and cloned into pPIC-ZA for expression. All amino acid substitutions were confirmed by DNA sequencing.
Arrestin was purified using two affinity matrices. The first utilized the N-terminal (His)6 tag of arrestin, binding to a Ni-agarose resin (HIS-Select HC, Sigma, St. Louis, MO). Arrestin eluted from the Ni-agarose by a 10-500 mM gradient of imidazole in 100 mM sodium phosphate with 300 mM NaCl (pH 8.0) was pooled, diluted with two volumes of 10 mM HEPES/15 mM NaCl (pH 7.0), and applied to a heparin-agarose column (Sigma). Highly purified arrestin was eluted from the heparin column using a 15-750 mM gradient of NaCl in 10 mM HEPES. Fractions containing the highest concentrations of arrestin were pooled and used for experimental assays.
Binding assays and antibody competition
Arrestin binding to rhodopsin in fragments of disc membranes and antibody competition assays were performed as previously described . Reaction mixtures contained 4 μM arrestin or myc-tagged arrestin mutant, 7 μM RP in disc membranes, and increasing concentrations of antibody (0-8 μM). The anti-myc antibody 9E10 (American Tissue Type Cultures, Manassas, VA) was affinity purified over protein A/G sepharose (ImmunoPure Protein A/G; Pierce, Rockford, IL) prior to use in competition assays.
Fluorescent labeling of arrestin
Cysteine substituted mutants of arrestin were fluorescently labeled with the maleimide derivative of AlexaFluor-594 (Molecular Probes, Eugene, OR). For this reaction, arrestin was immobilized on Ni-agarose via its N-terminal (His)6 tag and the AlexaFluor-594 continuously circulated over the arrestin in 10 mM HEPES/100 mM NaCl (pH 7.0) for 5 h at 4 °C. After rinsing to baseline, the labeled arrestin was eluted from the Ni-agarose with 500 mM imidazole in the mobile phase, and then dialyzed against 10 mM HEPES/100 mM NaCl.
For monitoring fluorescent emission of the labeled arrestin in the presence of dark and light activated phospho rhodopsin, labeled arrestin (10 μM) was mixed in the dark with sonicated membranes containing phosphorhodopsin (20 μM) prepared as described  and placed in a cuvette in a F900 fluorimeter (Edinburgh Instruments, Livingston, UK). Fluorescence was excited at 340 nm and emission measured at 550-660 nm with a 5 nm slit width. The ultraviolet excitation wavelength was selected to minimize the activation of rhodopsin as well as to minimize scattered light from sonicated membrane fragments. Following collection of the emission spectra, rhodopsin was activated to form R*P by illuminating the sample for 1 min with bright white light, and then remeasuring the emission spectra of arrestin bound to R*P. A difference spectra was calculated by digitally subtracting the emission obtained using RP from that obtained using R*P.
Myc tag insertional mutagenesis of loop structures in arrestin
Previous studies have suggested that the loop structures of arrestin are important in the binding interaction with R*P. For example, Dinculescu et al.  showed that elements in the 68-78 loop of arrestin appear to have a role in the direct interaction with R*P. Further, Pulvermuller et al.  identified several peptides spanning loop structures that were competitive inhibitors of the additional metarhodopsin II (extra meta-II) formed when arrestin binds metarhodopsin II. Accordingly, we used the loop structures to identify the surface of arrestin that binds to R*P. For this study, the ten amino acid myc tag (EQKLISEEDL) was placed in eleven different loops of arrestin (the nomenclature indicates the amino acids between which the tag was inserted in bovine visual arrestin; for example, 186myc187 has the myc tag inserted between Pro-186 and Gln-187). These tags were then used to competitively inhibit the binding of arrestin to R*P, using an anti-myc monoclonal antibody.
Figure 1A shows representative competition curves for the three classes of responses that were obtained: non-competitive (186myc187), weakly competitive (199myc200), and strongly competitive (250myc251). The IC50 was calculated from the dose-response curve for each of the tagged arrestins, including native arrestin without a myc tag (Figure 1B). These results show that the binding interaction between arrestin and R*P can be efficiently blocked when the myc tag is placed at Ile-72/Asp-73, Asp-162/Lys-163, Tyr-250/Ser-251, Leu-338/Leu-339, and Thr-343/Ser-344. All of these constructs were competitive at an IC50 <4 μM in a reaction containing 4 μM arrestin and 7 μM R*P. Significant competition was also obtained with the myc tag inserted between Ser-199/Asp-200, although the effect (IC50=8 μM) was not as great as that obtained for the above constructs. Positioning of the myc tag in the other loop structures of arrestin (i.e., 37myc38, 99myc100, 186myc187, 314myc315, and 399myc) were noncompetitive, with extrapolated IC50s all exceeding 15 μM. The results of this competition are mapped onto a representation of the three dimensional structure of arrestin (Figure 2). This diagram reveals that all of the competitive insertions map to one face of the arrestin molecule, whereas all of the noncompetitive insertion sites are on the opposite face.
Arrestin labeling with a fluorescent reporter
Although antibody competition is a well established technique for identifying interactions between proteins, it is not without limitations since it involves a large macromolecule interfering with the interaction of two other large molecules. Consequently, we also labeled arrestin with a fluorescent reporter molecule to determine if a particular loop structure was in close proximity to R*P in the bound complex. For this labeling, single cysteines were introduced at Ile-72, Val-177, and Ser-251, and double cysteines were substituted for Ser-156 and Ser-199 (Figure 3A). The positions of the cysteines were selected on the basis of occupying regions predicted to have a tight interaction with R*P (I72C and S251C), a competitive but indeterminate interaction with R*P (S156C/S199C), or a region that is not predicted to interact with R*P (V177C).
These molecules and native arrestin were reacted with AlexaFluor-594 maleimide. Excitation of the reacted proteins with ultraviolet light shows that each of the proteins is labeled by the fluorophore (Figure 3B), although I72C, S156C/S199C, and S251C were labeled more completely. The poor labeling of native arrestin, which has three free cysteines, and V177C reflects the inaccessible nature of these cysteines [17,20], compared to the other cysteine substitutions that were designed to be on the surface of the arrestin molecule. These labeled arrestins all retain their binding selectivity for R*P (Figure 4).
AlexaFluor-594 is a fluorescent molecule whose emission efficiency is sensitive to its environment, with greater emission intensity in a more hydrophobic environment than when surrounded by an aqueous environment . The fluorescent emission of these labeled arrestins was measured, comparing emission intensity of unbound arrestin to arrestin bound to R*P (Figure 5). These results show an increase in fluorescence emission for the labeled I72C and S251C proteins, whereas V177C and S156C/S199C proteins showed no change in emission upon binding to R*P. These results are consistent with the results obtained from antibody competition (see Figure 1), indicating that the 68-78 loop and the 248-254 loop are in close proximity to R*P in the bound complex. We note that there is a difference in the absorption maximum for I72C (618 nm) compared to S251C (605 nm). Although we have no definitive explanation for this difference, we suspect that the abundance of aromatic amino acids (tyrosine and phenylalanine) in the vicinity of Ser-251 may alter the spectral emission of the fluorophore. The lack of increased fluorescence for S156C/S199C, even though they are in close proximity to loops that were shown by antibody competition to interact with R*P, would suggest that these loops of arrestin are not closely apposed to R*P in the bound complex.
Scanning mutagenesis of 248-254
Since our antibody competition and fluorescent reporter studies indicated that the 248-253 loop is likely involved in binding to R*P, we conducted scanning alanine mutagenesis for the region 248-254. Figure 6 shows that alanine substitutions of Val-248, Asp-253, and Tyr-254 did not significantly impact the relative binding selectivity for R*P. In contrast, substitutions of residues 249-252 decreased the relative binding to R*P by approximately 20%, with maximum effect on binding obtained for alanine substitutions at Ser-251 and Ser-252.
The principal finding of our results is that we empirically demonstrate the surface of arrestin that interacts with R*P. Figure 2 shows all the tagged loops that can be used to competitively block binding to R*P lie along a single surface encompassing the N- and C-terminal domains. This surface has been previously proposed to interact with rhodopsin [13,22,23], although direct evidence is limited. Our study provides convincing experimental evidence of this interaction and helps delineate the extent of the binding matrix involved in arrestin binding to rhodopsin. Beyond identifying the surface of arrestin that binds to rhodopsin, our results also suggest that the surface around loops 68-78 and 248-253 may be near the center of the bound complex. Fluorescent labeling of either loop at Ile-72 or Ser-251 results in increased fluorescent emission of the fluorophore when arrestin was bound to R*P, indicating that the environment of the fluorophore becomes less solvent exposed upon binding. This finding is consistent with the fluorophore either moving into close contact with rhodopsin in disc membranes or with the fluorophore being moved to a more buried position within the arrestin molecule. Because our myc competition and fluorescent labeling studies both showed that loop 248-253 was near R*P in the bound complex, we performed site-directed scanning alanine mutagenesis to determine if any residues in this loop might bind directly to R*P. This analysis showed that alanine mutagenesis of Leu-249, Tyr-250, Ser-251, and Ser-252 each caused a relative decrease in binding by about 20%, with S251A and S252A being the most influential. The lack of a dramatic effect on binding to R*P suggests two interpretations. First, Ser-251 and/or Ser-252 may bind directly to R*P, but form just one of several contact points between arrestin and R*P. Alternatively, the scanning mutagenesis may cause shifts in the side chains of closely apposed amino acids that are directly involved in binding to R*P.
There are other studies in the literature that support the involvement of the 248-253 loop in binding selectivity for R*P. Hofmann's research group looked at the ability of overlapping 20-mer peptides of arrestin to compete for the stabilization of extra-metarhodopsin II formed by the binding of arrestin to R*P . In this study, three peptides that were effective at competing for arrestin stabilization of extra-metaII were peptides 231-250, 231-260, and 241-260. These peptides all include at least a portion of the 248-253 loop of arrestin. Peptides flanking this loop (221-240 and 251-270) were ineffective competitors. Similarly, serial truncations and domain swapping between visual arrestin and β-arrestin have implicated broad regions in the C-terminal domain of arrestin (that include the 248-253 loop) to be involved in the high affinity binding of arrestin to rhodopsin [23,24]. This loop was also contained in a region of arrestin that was identified to bind to R*P using phage display of arrestin fragments . These studies all support the involvement of loop 248-253 in binding to R*P, although they were not performed at the same degree of resolution as the current study.
As stated previously, ours is not the first study to propose that multiple sites on arrestin are involved its binding to R*P, and in fact this is a common theme in the study of arrestin/rhodopsin interactions. In the study of arrestin, two larger questions remain unanswered. The first is how does arrestin rearrange itself to bring together the molecular elements that selectively bind to R*P? Are these intramolecular changes small, involving changes in loops as some studies have suggested , or are they much larger, involving movements of the N- and C-terminal domains [4,12,14]. The second large question is what is the oligomeric state of both arrestin and rhodopsin in the bound complex? Several studies have suggested that at physiological concentrations, arrestin should exist as at least a dimer in rods [15,16,26]. And not only arrestin, but rhodopsin may also exist as a dimer in disc membranes [27,28]. It is easy to conceptualize how the larger arrestin (approximately 45 nm2 on its N- and C-terminal domain surface) could bind to more than one rhodopsin molecule that has a much smaller cytoplasmic surface (approximately 15 nm2). However, the likelihood of adjacent molecules of rhodopsin absorbing a photon at illumination intensities used by rods is vanishingly small. Clearly, both questions will be resolved upon solution of the arrestin/rhodopsin three dimensional complex.
This research was supported by grants (EY06225, EY06226, EY08571) from the National Eye Institute, and by grants from Research to Prevent Blindness (RPB).
1. Burns ME, Baylor DA. Activation, deactivation, and adaptation in vertebrate photoreceptor cells. Annu Rev Neurosci 2001; 24:779-805.
2. Arshavsky VY, Lamb TD, Pugh EN Jr. G proteins and phototransduction. Annu Rev Physiol 2002; 64:153-87.
3. Gibson SK, Parkes JH, Liebman PA. Phosphorylation modulates the affinity of light-activated rhodopsin for G protein and arrestin. Biochemistry 2000; 39:5738-49.
4. Vishnivetskiy SA, Hirsch JA, Velez MG, Gurevich YV, Gurevich VV. Transition of arrestin into the active receptor-binding state requires an extended interdomain hinge. J Biol Chem 2002; 277:43961-7.
5. Schleicher A, Kuhn H, Hofmann KP. Kinetics, binding constant, and activation energy of the 48-kDa protein-rhodopsin complex by extra-metarhodopsin II. Biochemistry 1989; 28:1770-5.
6. Palczewski K, Pulvermuller A, Buczylko J, Hofmann KP. Phosphorylated rhodopsin and heparin induce similar conformational changes in arrestin. J Biol Chem 1991; 266:18649-54.
7. Puig J, Arendt A, Tomson FL, Abdulaeva G, Miller R, Hargrave PA, McDowell JH. Synthetic phosphopeptide from rhodopsin sequence induces retinal arrestin binding to photoactivated unphosphorylated rhodopsin. FEBS Lett 1995; 362:185-8.
8. Raman D, Osawa S, Weiss ER. Binding of arrestin to cytoplasmic loop mutants of bovine rhodopsin. Biochemistry 1999; 38:5117-23.
9. Raman D, Osawa S, Gurevich VV, Weiss ER. The interaction with the cytoplasmic loops of rhodopsin plays a crucial role in arrestin activation and binding. J Neurochem 2003; 84:1040-50.
10. Dinculescu A, McDowell JH, Amici SA, Dugger DR, Richards N, Hargrave PA, Smith WC. Insertional mutagenesis and immunochemical analysis of visual arrestin interaction with rhodopsin. J Biol Chem 2002; 277:11703-8.
11. Pulvermuller A, Schroder K, Fischer T, Hofmann KP. Interactions of metarhodopsin II. Arrestin peptides compete with arrestin and transducin. J Biol Chem 2000; 275:37679-85.
12. Vishnivetskiy SA, Schubert C, Climaco GC, Gurevich YV, Velez MG, Gurevich VV. An additional phosphate-binding element in arrestin molecule. Implications for the mechanism of arrestin activation. J Biol Chem 2000; 275:41049-57.
13. Han M, Gurevich VV, Vishnivetskiy SA, Sigler PB, Schubert C. Crystal structure of beta-arrestin at 1.9 A: possible mechanism of receptor binding and membrane Translocation. Structure (Camb) 2001; 9:869-80.
14. Vishnivetskiy SA, Hosey MM, Benovic JL, Gurevich VV. Mapping the arrestin-receptor interface. Structural elements responsible for receptor specificity of arrestin proteins. J Biol Chem 2004; 279:1262-8.
15. Shilton BH, McDowell JH, Smith WC, Hargrave PA. The solution structure and activation of visual arrestin studied by small-angle X-ray scattering. Eur J Biochem 2002; 269:3801-9.
16. Imamoto Y, Tamura C, Kamikubo H, Kataoka M. Concentration-dependent tetramerization of bovine visual arrestin. Biophys J 2003; 85:1186-95.
17. McDowell JH, Smith WC, Miller RL, Popp MP, Arendt A, Abdulaeva G, Hargrave PA. Sulfhydryl reactivity demonstrates different conformational states for arrestin, arrestin activated by a synthetic phosphopeptide, and constitutively active arrestin. Biochemistry 1999; 38:6119-25.
18. Smith AM, Klugman KP. "Megaprimer" method of PCR-based mutagenesis: the concentration of megaprimer is a critical factor. Biotechniques 1997; 22:438,442.
19. McDowell JH. Preparing rod outer segment membranes, regenerating rhodopsin, and determining rhodopsin concentration. Methods in Neurosciences 1993; 15:123-30.
20. Palczewski K, Riazance-Lawrence JH, Johnson WC Jr. Structural properties of arrestin studied by chemical modification and circular dichroism. Biochemistry 1992; 31:3902-6.
21. Imamoto Y, Kataoka M, Tokunaga F, Palczewski K. Light-induced conformational changes of rhodopsin probed by fluorescent alexa594 immobilized on the cytoplasmic surface. Biochemistry 2000; 39:15225-33.
22. Granzin J, Wilden U, Choe HW, Labahn J, Krafft B, Buldt G. X-ray crystal structure of arrestin from bovine rod outer segments. Nature 1998; 391:918-21.
23. Gurevich VV, Benovic JL. Visual arrestin interaction with rhodopsin. Sequential multisite binding ensures strict selectivity toward light-activated phosphorylated rhodopsin. J Biol Chem 1993; 268:11628-38.
24. Gurevich VV, Dion SB, Onorato JJ, Ptasienski J, Kim CM, Sterne-Marr R, Hosey MM, Benovic JL. Arrestin interactions with G protein-coupled receptors. Direct binding studies of wild type and mutant arrestins with rhodopsin, beta 2-adrenergic, and m2 muscarinic cholinergic receptors. J Biol Chem 1995; 270:720-31.
25. Smith WC, McDowell JH, Dugger DR, Miller R, Arendt A, Popp MP, Hargrave PA. Identification of regions of arrestin that bind to rhodopsin. Biochemistry 1999; 38:2752-61.
26. Schubert C, Hirsch JA, Gurevich VV, Engelman DM, Sigler PB, Fleming KG. Visual arrestin activity may be regulated by self-association. J Biol Chem 1999; 274:21186-90.
27. Liang Y, Fotiadis D, Filipek S, Saperstein DA, Palczewski K, Engel A. Organization of the G protein-coupled receptors rhodopsin and opsin in native membranes. J Biol Chem 2003; 278:21655-62.
28. Fotiadis D, Liang Y, Filipek S, Saperstein DA, Engel A, Palczewski K. Atomic-force microscopy: Rhodopsin dimers in native disc membranes. Nature 2003; 421:127-8.